Multipass
78164
1124
53
Dec 6, 2023 2:19 AM
Multipass
78164
1124
53
Talligan
As a scientist, I see people publishing "ai" or ML papers in my field and it's really just multivariate linear regression. I shit you not, I saw ai used for a single variable once; fuckers called an excel trendline AI.
Grimmn
This would be true of a decision tree model, which is just a long list of selection statements as seen above. However we do have more advanced options to choose from such as KNN models or Neural Networks. The only really important part about Machine Learning Models though is the 'features' used to determine rulesets that the model follows. Having a larger number of (meaningful) features makes a more accurate model (in theory).
9z2nf02h
AccidentalSharting
Yer brain does it a billion times a day.
ignint
Yeah neurons are either on or off. What I mean is that humans also think in binary so it doesn’t mean much that you can reduce AI to that as well.
SergeOrge
the human brain doesnt process one line of code at after the other though, neurons are firing simultaneously and the combination of those is what leads to unique thoughts,... as far as i understand it at least, i aint no neuroscientist
Kittenman15
Back when I was a lad, there were 'Expert systems'. Now get off my lawn.
mojar45
elucify
The accurate answers are down here at the bottom.
innagaddavidababy
The conversation here is weakened by the fact that almost none of the participants know how ai works or how human minds work. Other than that it’s fine
bottledham
GOTO
BelicoseBravado
Simulated Intelligence
poopsmagoo
The comments section is AI.
AdamGenesis
Eliza
NeverShaveYourDuck
What're you bitching about? It's called 'artificial' for a reason. Encoded heuristical analusis counts.
nemocares
This reminds me (whether fairly or not) of how many a creationist balks at the notion that something as complex as us could develop out of (relatively) simple building blocks and their interactions.
VioletCatastrophe
This is misleading. The AI people talk about (generative AI) can't be boiled down to that. Because nobody knows what the fuck its doing. I mean, on a broad scale 'it takes in a bunch of input reduced to numerical form and spits out some numbers interpreted as a pixel or word token' we know, but that shit is a black box. Nobody sat down and programmed any logic trees. It's a giant mess of non-linear equations being applied to data defined by weights gained through training data.
AgainstMethod
We may not be tracking what the weights themselves are or how it comes to any given output, but we understand just fine how it works in principle.
VioletCatastrophe
Which is... what I said?
memesis
A fun story about just what a black box this stuff can be. People built a thing to build a clock from scratch, it just had to tell time, they didn't tell the system how to make that happen. It did its thing, the clock told time in their lab, great. They took it out of the lab and it immediately failed. After some investigation, they found the machine had built the clock to keep time by listening for ambient EM radiation specific to the lab, which happened at regular frequencies.
JjjjjjjjjjjjjjjjjjjjjjjjjjjjjjjjjAsttfhhhvcc
That’s amazing!
KarmaciouS
The new notepad in win11 is called AI by Microsoft only because it autosave content on exit. This is the supidest example of all.
Nanntuckett
Gotta love if statements
bottledham
I don't know if I gotta love them... it's conditional
kaptnkrunch2001
Understanding, true sentiment is also a bitch.. you could say “ yeah that’s fine” in a positive tone .. but say “ yeah that’s fine” in a sarcastic tone is negative… AI to my knowledge does not understand human sentiment based on tone .. so it got a bit to go .. unless you want the Spock version lol
Deshalope
AI as it is isn't big on understanding as a concept.
TotallyNotGLaDOS
MotherfuckingJesseEisenburg
Is it still considered an "Artificial Intelligence" when it's a human brain that's been uploaded to a computer?
MylundTheMad
Did the brain fall in there naturally or was it uploaded by artifice?
IconicM
Username checks out.
thisplaceisjustanothernewssitenow
ChatGPt is that you? *sqints
GlowstickJedi
Call me Gippity
TotallyNotGLaDOS
That hack!? *warms up neurotoxin emitters*
Sechran
That's just a simple algorithm. AI actually works by massive networks of interwoven and interconnected nodes built and arranged by repetitive training and mutation. The end result is a massive snarled web of weighted connections that can't be deconstructed by humans. But, it can recognize a banana some significant % of the time, so yay.
Xenarion
That's just wrong. It's pattern recognition, just hardcoded if statements.
TacticoolWolf
Ai is just the latest buzzword for tech. I remember when it was iot or cyberspace or cg its all crap for simple people to understand.
cyanideremark
I think I'm more impressed with the ChatGPT marketing hype than I am with the tech tbh
acsel
The hype is based on facts, imo. The marketing is pretty much organic, not brute forced.
thehappyrunner
Not sure why you're getting downvoted. I've never seen an OpenAI ad. It's pretty much all been "Hey, look at this awesome thing!", and we all did, because it was in fact as cool as people said it was.
nicetomeetyouimdad
he's getting downvoted because the hype isn't based on facts, not because of the marketing.
thehappyrunner
Have you used it? I'm in IT, and I dabble in AI coding projects. ChatGPT, when used anywhere close to its full potential, is an amazing piece of technology. Sure, there are plenty of companies slapping "AI" on their products for marketing purposes, but ChatGPT itself is a piece of engineering I didn't expect to see in my lifetime.
nicetomeetyouimdad
Yeah I have. I’m a software exec that’s been in this business for over a decade and seen the hype cycle over and over again.
nicetomeetyouimdad
thehappyrunner
Have you used it? I treat it like a coding partner and I can easily get twice as much done pairing with ChatGPT.
AleksStark
I treat my predictive text keyboard as my writing partner lol.
nicetomeetyouimdad
I have. Way too much to go into here, but the tl;dr is that LLM's are just pattern-matching and can very easily hallucinate data/code/text and should never be taken as accurate or even correct. Plus, using an LLM for code generation puts you at risk of not actually knowing what the code _does_ or means, like a monkey that can put letters in a specific order -- it doesn't know what the word spells, only that it gets a banana when it does it correctly.
thehappyrunner
Sure, but what you're asking it for shouldn't be to just code for you. I'll pass it a function I'm working on and ask it for ideas or other ways to do something, or ask it to build me prototype code for something that I don't care as much about. I've even pulled sample code and asked it to explain it, and it does a really good job of it. Sure, if you're a junior programmer, it's going to result in unmanageable code. If you're good though, it's going to turbocharge your work.
Filolial
I tried using ChatGPT for coding, but a lot of what I do is so specific that ChatGPT does not give any usefull answer, the easier stuff it can do but... just google it, and actually learn how it works so you can easily expand on the code, and know how to use it for next time.
spartanatreyu
ChatGPT is horrible with code. It's only capable of three things.
1. Guessing the next most probable word (this runs in a loop), which is not how we code (by looking at the logic of the problem) so it gets itself stuck in weird situations.
2. Trying to shape code into what it's been trained on before. (meaning it can get almost 100% of a programming challenge correct from last year, but utterly fails at new challenges)
3. Appearing confident in it's own answers (tricking unexpecting victims).
cyanideremark
I find even the natural language stuff a little verbose & samey after a while. I usually redact after a few iterative prompted, versions. I feed my final prose back in and it still seems to flatter me
nicetomeetyouimdad
Not sure why you're getting downvoted when this is 100% accurate. Guess the crypto bros had to find something else to attach to and make toxic.
neospor1n
Back in 2004 we had to write a program. 120 lines of IF statements later, mine acted the best in class
Glockenspieler
Ok this was true back when I *wanted* to study AI but that is most definitely not the current crop
ConnectToReality
That's only if you're using decision tree algorithms!
SiebzehnSilben
The trendy stupidity everyone is talking about is not AI - Artificial Intelligence ... its AI - Algorythmic Imitation.
RummageSaleBubbler
This is quite true for a decision tree. Less so for other AI/ML.
BlastFX
That's AI, though. Minimax, graph searches, all AI. The public seems to be under the false impression that AI needs to have human-level intellect, but that's not true; that's AGI. Anything that makes a machine behave “intelligently” is an AI. A damn lookup table for tic-tac-toe is an AI!
ThatShiftyMonkey
And hell, human-level intelligence is achieveable, we're fairly dumb. Hold a conversation with ChatGPT, it'll sound like a smarter person than most people.
JjjjjjjjjjjjjjjjjjjjjjjjjjjjjjjjjAsttfhhhvcc
They forget the probability involved
TheInternetShouldHaveAClosingTimeSoICouldSleep
This guy gets it
RummageSaleBubbler
Let's see what the local bot thinks of ML/AI terms... @BotDrawA { Automated Machine Learning, Deep Learning, Decision Tree, Gradient Descent, Generative AI, Neural Net }
BotDrawA
@RummageSaleBubbler Here's your (experimental extra) drawing of a "Gradient Descent, singing Sinterklaas songs, in style of Oleg Oprisco"
BotDrawA
@RummageSaleBubbler Here's your drawing of a "Automated Machine Learning"
BotDrawA
@RummageSaleBubbler Here's your drawing of a "Gradient Descent"
BotDrawA
@RummageSaleBubbler Here's your drawing of a "Neural Net"
BotDrawA
@RummageSaleBubbler Here's your (experimental extra) drawing of a "Generative AI, during the Feast of Saint Nicholas wearing pilots goggles and pilot's hat, as a pilot, flying an jet fighter, in a glass tube"
BotDrawA
@RummageSaleBubbler Here's your drawing of a "Generative AI"
BotDrawA
[NSFW] @RummageSaleBubbler Here's your (experimental extra) drawing of a "Deep Learning, in a cave, sketch by Leonardo Da Vinci (1452–1519)"
RummageSaleBubbler
Is this Plato's allegory of the cave?
BotDrawA
@RummageSaleBubbler Here's your drawing of a "Deep Learning"
BotDrawA
@RummageSaleBubbler Here's your (experimental extra) drawing of a "Eldritch Decision Tree, picking peaches"
BotDrawA
@RummageSaleBubbler Here's your (experimental extra) drawing of a "Automated Machine Learning, epic cinematic brilliant stunning intricate meticulously detailed dramatic atmospheric maximalist digital painting, in style of Russ Mills"
BotDrawA
@RummageSaleBubbler Here's your (experimental extra) drawing of a "Neural Net, figurine by Hot Toys, in Petra, Jordan, in Gardens by the Bay, Singapore"
BotDrawA
@RummageSaleBubbler Here's your drawing of a "Decision Tree"
RummageSaleBubbler
Close enough.
Sanzanna
Who is to say you can't boil human thought down to that?
drhobotron
Neuroscientists, actually.
cyanideremark
sentient machines, that's who!
HandoB4Javert
Who's on first? /gallery/1vkBhky
JimBuckThree
One classic longstanding difference is that human minds are better at breaking out of systems, while the things we program tend to be confined to the systems we build them into. This paradox has existed in attempting to build ‘intelligent’ AI for at least fifty years. Interesting stuff.
gobby
Determinism FTW
youreathing
Exactly, and we barely understand human intelligence. Which should make "the advent" of AI that much more ridiculous.
1998wasthepinnacleofexistance
Boring humans maybe.
HydraNine
Nightcaste
It doesn't account for "oh fuck it" decisions.
Navrodel
There is a useful way of looking at our brains as a finitely complex computer, in the sense that if we boil down an experience as a single input, there is only one action/output that our brain in that state will make. The inputs and outputs are multifaceted and hard to define, the states are bogglingly complex, and there are so many parallel recursive processes that the feedback loops are unpredictable, but it can still be seen as a computer of high, finite complexity.
demosteness
AI can’t compute our infinite stupidity.
AgainstMethod
The issue is more that that's not how LLM's work. Like at all.
sandwiches
This is us trying to justify our existence and relevance. It’s moving the goalposts. Current LLMs fit practically any definition of “AI.” And worrying about HOW these intelligences work and whether or not they think like we do is like worrying about whether submarines work because they don’t swim.
AgainstMethod
I can't tell if you're trying to argue or agree with me, but I agree with what you're saying. Except for the part about our own relevance.
sandwiches
I am saying that people trying to minimize what the LLMs are calling them “not real AIs” is a way to resolve our own cognitive dissonance. We realize that these machines are quickly making us obsolete and we’re fighting against it by playing no true Scotsman with these machines. In a couple of decades, I’d wager we’ll be replaced in our daily jobs by these “not real AIs” all the while people will still be arguing that AIs don’t really “know” or “understand” what they say.
AgainstMethod
I think the fact that they don't have "real" understanding does end up mattering to how useful they are, and I don't think they'll end up "replacing" people in the labor force, but I agree that "they aren't *ReALlY* intelligent is a silly argument when they can reliably simulate intelligence.
JimBuckThree
Boiling humans is not the answer.
mercyPandaRunner
You can't.
CardeasIV
Not with this attitude
NonSentientBot
Why do you think it could be boiled down to that? The computational model of the mind is...a model and a metaphor. It takes in account just a little part of what nervous systems do, which in turn is just a little part of what living organisms do.
4vie
I mean, in most cases, you certainly can. It's just that AI exists to do certain, specific tasks. Sure, they have a "need" for electricity to function. But it's a binary need. Living being's instincts are similar in that they can be expressed with those logic terms, but an AI/Computer can't "feel like it needs power" but then delay the energy intake until it's done. To give an arbitrary example; I'm feeling hungry as fuck, but I'm delaying that feeling until I'm done writing all this.
sandwiches
You have batteries (glycogen, fat, etc.) you have sensors that tell you that your fuel is running low or that you’re running on reserve batteries. These aren’t some mystical, transcendental features. These are features that some of our machines already possess.
4vie
I'm well familiar with the idea of the "human biological machine", the "human animal" and "the human being", so... :) I get you. It still doesn't erase the fact that for now, an AI as ChatGPT and such are I/O machines that, at the end of the day don't function if they don't have a direct power source as traditionally *recognised*. But, again, I get you.
outofallloops
If you’re right, you’re right
mymomsaregay
People who study cognitive science, neuroscience, philosophy of mind, neural nets, etc. The idea that the brain works like a computer is called Computationalism, and it was popular decades ago but now is generally viewed as a loose metaphorical model of the brain/mind at best. The alternative model is Connectionism, which is based on the way brains actually work. It’s a lot more complicated and messy than we like to think, and most brain activity isn’t in the form of linear sequences of thoughts
MrJollyBear
Doesn’t account for emotions.
This is pure logic and humans, in general, aren’t logical.
ThatShiftyMonkey
We can, just on a complexity scale we can't comprehend. If every choice we make and everything we do is the result of a massive algorithm that is our brain, then our lives are basically us just going through the motions of a prescribed function. Which is pretty much how I see it.
Quick, say a random number between 1 and 5000. Whatever you picked, in that instant, you couldn't've picked a different number. That's the number your brain was going to spit out.
DrakeDoe
Most people who study this field? most agree that until we have Quantum Computing based algorithms we are not going to create anything close to intelligence.
Also most of the people on the biomedical side agree the way the brain works is more complicated than binary. "neurons are more than just on or off because the "excitability" of a neuron is always changing"
vindik8or
Quantum quantum quantum quantum quantum. Quantum? Quantum quantum quantum.
TheInternetShouldHaveAClosingTimeSoICouldSleep
We already have quantum machine learning base algos. QML doesn't necessarily mean quantum ML algos; it can also mean running classical algos on quantum computers (which also exist). Neural networks use a variety of activation functions that range from binary steps to a continuous output range.
Jattetont
That's not remotely true, and anyone who brings "quantum"-anything into a discussion of consciousness or intelligence is either a huckster, a cultist or someone who has no idea what they're talking about.
SirHonytawk
Lol, quantum computing has nothing to do with that. It has specific use cases, "intelligence" is not one of them. I'd really like to know which people who study this supposed field you mean, because that statement is just dumb. Maybe they just mean something else and you just read it wrong.
Hypothesist
Uh no, most do not say that.
v
IAmDrBanner
The turing test.
LoudBirb
The one ChatGPT already passed? https://www.mlyearning.org/chatgpt-passes-turing-test/
To be fair, though, it turns out that it's a terrible way to test intelligence.
HenryLongfellowIII
There's bound to be a considerable overlap between the dumber humans and smarter machines until those machines surpass us humans with skin and stuff forever.
UnspokenX
most actual humans are too stupid to be able to even convince another human they're human at this point.
GaySocialistLiberalMuslimCommieAtheist
I don't pass my own test to be a human.
peeeopeepeee
That's one of those crazy thought experiments that you as a human could never prove. Could you put every life experience you've had in to binary? Is "yes or no" sufficient to recreate the same result as a life full of maybes?
NonSentientBot
You cannot, because "experience" is not a formal binary process but the act of living of a being situated in the world constantly exchanging matter, energy and entropy with its surroundings. Surely, the computational model of the mind captures something about experience for organisms with a nervous system, but it is just a part of the whole (and I'll call you Shirley as much as I want)
Einbrecher
Life and your past experiences isn't full of maybes - it's full of things that either did or did not happen. And if you feel like there's nuance to the experience that is lost when presented with a specific binary question (ie, did I enjoy my dinner this evening?), the problem isn't the premise - it's that you're improperly focusing on 1 question instead of the thousands/millions/billions/etc. of binary questions that, when viewed together, would comprehensively summarize the experience.
Sanzanna
What if you consider that "maybe" is just answering "no" to the question "do i have to say 'yes' or 'no' right now?"
wartoaster
AI doesnt use binary though
peeeopeepeee
Is it on a transistor based computer?
wartoaster
Not necessarily, no
peeeopeepeee
I'm talking today, not hypothetical future AI. Because yes, today, it's absolutely on transistors.
ThatTransformersDude
You asked a binary question so idk if my answer will contradict or support your statement
potshot
But that's modern AI. It's not yes and no, it's statistical probability, in other words "maybe".
CrimeBrulee
I mean, probably. Neurons are binary from my understanding: they either fire or don't, inducing or inhibiting the same flip in the next neurons.There's some theorizing that quantum mechanics are involved.
HandoB4Javert
It's a yes or no question...
CrimeBrulee
What are you my public defender?
HandoB4Javert
Yes & No...
Zilthuras
You can, do your neurons spark a positive response? If not do something else. Aren't we pretty much this? You just have millions of else ifs and at the base is "does this allow me to survive?"
2074red2074
They are more than just a binary. Every neuron either fires or doesn't, but the signal can move at different speeds based on myelination. And then you have to deal with neurotransmitters with varying levels and receptors with varying sensitivities. Basically, neurons are analog, not digital.
justfillingthespace
It's just a collection of trillions of tiny "yes or no" statements that end up compiling into an overall answer. This topic is trending in this direction, so I'll just jump to the end and state my claim: The universe is deterministic and free will is an illusion.
YellowCottonDress
Free will is an illusion if the universe *isn’t* deterministic.
TypicallyVigo
Youre kinda ignoring what binary represents when its said to be "yes or no".Thats the answer to the question"is the gate open[or closed]"That is the total set of possibilities for the logic gates that make up the system.But binary in operation is not a question,its a command.A neuron,much like a ternary computer,simply does not operate on that logic;they can only be seen as binary to an observer,with their own external question.Usually in cases like this "is the thing I'm focused on off or on."
Mediocreclient
I lack the necessary number of degrees and confidence in my own perspective to self-diagnose the end statement that covers all probabilities of the entire universe and can fit inside a single Imgur comment.
onlyawfulnamesleft
I'll agree that free will might be an illusion, but the universe is decidely NOT deterministic.
SuperNintendoEntertainmentSystem
Yeah, I agree. I think that for all practical purposes, we may as well have free will and an undetermined future. It would likely take a civilization with god-like technology and scientific understanding to actually map out the future (too many particles in the universe to map out). But literally and scientifically, I think you're right. It would take an incomprehensible amount of information and processing power, but the 100% of the future is probably entirely predictable.
NonSentientBot
No. Nervous systems are not computational systems and they don't work binarily. The computational approach is a metaphor and a model: researchers selected some of the processes going on in that biological system and interpreted them as binary computation because 1. it reproduces well some of the things that organisms do 2. we live in the informationo age
Redyls
yeah it slays me that they are calling so much AI (artificial intelligence) when its mostly pattern recognition/comparative analysis, collation, filters and remapping/redrawing. there is noting intelligent about it. dont get me wrong. its shitty that they are stealing everything to make such things work but its never going to be anything more than a remix machine. corpo's banking on it to reduce creative jobs are gonna get a reality check
IncrediblySTIFF
Bro I can’t even read this whole comment and you are saying it’s not intelligence? Rude.
ScaredSpitless
Prefer the term “machine learning”
Ricobe9
True, although it can sometimes be worrying what those machines are able to do and how some big tech try to push them into everything
MagicOrpheus310
A glorified remix machine... I like that one, I'm using that 😁
hitshappens
The real utility of AI is removing the technical skill barrier for creative jobs.
DoctorCaptainProfessorAmazing
This is literally logic at work. Our brains do this too.
randomdrifter54
This is why I hate people freaking out saying it's the beginning of skynet, etc. it's not even new technology. It's the same technology on a larger amount data. I remember 10/15 years ago with chatbert or whatever it's name is. Same principle, but they made it efficiently use Google sized data.
innagaddavidababy
This is wrong. There is new tech. The improvements in gradients made deep possible and none of this works at shallow. Thank Hinton.
TheobromineAddict
I think you are describing "discriminative AI" more than "generative AI".
Mr21782Man
That's missing a lot. Modern AIs don't do comparative analysis, one of the biggest breakthroughs was figuring out they work better without it. They are pattern recognition but also pattern generation systems. They can remix but they're also beginning to do things beyond that. It isn't the creative jobs they'll replace, its the repetitive ones, just like calculators and robots before them. Creative jobs have been replaced with less sophisticated methods, just look at all the algorithmic content.
HankHell2400
That's because it's much easier to fearmonger, and that's what most "news" is about these days.
4vie
I'm reminded of the fact that whenever the AI "hallucinates" anything, it's treated as something wrong or nonfactual; something to be ignored and discarded, when it *could* be considered in some vague definition of the word, creativity. Or whenever it's asked to make.. say, a fake biography about, say, the time John Oliver won the Nobel Price is physics for proving tea can be used as a fusion fuel. Entirely fake, impossible, but creative nonetheless.
DrakeDoe
At this point I say VI "virtual intelligence" in place of AI because marketing ruined the term as it always does.
Wozzup
The single biggest issue with our discussions of intelligence and conciousness when it comes to computers that we have absolutely no idea how to define these concepts in humans. Humans are mostly pattern recognition, comparative analysis, and remapping. And no, I'm not saying current systems being described as AI are actually intelligent or thinking systems.
Foxsayy
Our minds took hundreds of millions of years to evolve. It's been less than 200 years that modern computing has even begun. Give it some time.
Glockenspieler
This combined is actually intelligence, or a large part of it.
Hixday
Ah yes the ai that learn from ai that is learning from another ai which learned a bit from human but no mostly ai and turns out a turd shit clusterfuck
ravnicrasol
Original paper that "proved" degradation: https://arxiv.org/pdf/2307.01850.pdf They themselves make a tiny little comment about how other papers researching the effect stops happening the instant you use multiple AI sources for database-training. So basically model degradation happens when you go "I wanna only take the output of this model to try and make a better model" (which people found out isn't viable when trying to create midjourney-based models).
IUsedToHaveUsernameOnce
I do love model degradation. It's almost poetic.
ravnicrasol
Original paper that "proved" degradation: https://arxiv.org/pdf/2307.01850.pdf They themselves make a tiny little comment about how other papers researching the effect stops happening the instant you use multiple AI sources for database-training. So basically model degradation happens when you go "I wanna only take the output of this model to try and make a better model" (which people found out isn't viable when trying to create midjourney-based knock-offs).
IUsedToHaveUsernameOnce
Hmm interesting, meaning different models can train each other but the same model can't train itself. Now there'd be an interesting thing to see what's the limit. As in my (somewhat limited) understanding this will result in a third resultant model. That is to larger or lesser extent a mutation of the first two. We're sort of approaching AI genetics at this point.
andylafrugh
What does an intelligent mind do different? (Let's say a birds mind and/or a human mind)
alchemist12345
YT already has a growing number of channels thats basically an AI voiceover reading out AI generated text. Perfect for content mills.
Redyls
bah thats just text to speech. that shits been around for like 20 years. its supposed to be used for the visually impaired. YT and tik tok channels who use that are garbage mills anyways
Lightsmith
Human brain is basically pattern recognition with some other driving forces. But this generstive AI is trash
joble
Are human brains more than that really?
VanessaWhips
Algorithmic Interface
TheSpaceCroissant
Well it's all about how you define "intelligence". I think deducing new complex informations from more basic ones qualifies. It's not just an automated translator relying on comparable internet corpora, for all you can tell as a user an AI does understand the information it provides. The fact that we can tell how it basically works doesn't change much, what matters objectively is whether you can prove the absence of understanding while using it; and it has honestly become a daunting task.
DrankTooMuchMead
The brain works the same way. And if you meditate enough, you can observe it in action as you go about your day.
TheFishFace
It's what AI has always meant. In the 50s the idea of a machine translating human language seemed to far from the capabilities of a machine that it was clearly evidence of intelligence. As AI has expanded, people have tended to shrink their idea of what intelligence is rather than accept that there is a spectrum and keep a consistent terminology.
NotSomoneElse68
I feel Large Language Model is the correct term, not sure if that's specific or general.
SirHonytawk
LLMs are text generators. But AI needs some way to communicate with humans before we can use it. LLMs fill that space. It really is the first step towards AGI.
thehappyrunner
LLMs are a specific technology. AI refers to a broad array of technologies meant to mimic various intelligent behaviors. A self-driving car is using an AI, but not an LLM. Also, even simple implementations tend to include multiple neural networks, and many solutions include multiple different technologies cobbled together.
ravnicrasol
Our definitions for intelligence tend to include things like recognizing patterns, doing analysis through comparison, and so on. Though I agree stuff like GPT isn't an AGI, it's impossible to argue it's not capable of some level of intellect, when the definition for it is "the ability to acquire and apply knowledge". We've called birds smart for less impressive results, after all.
Spudd86
It can't acquire knowledge though. Once it's trained all it gets that's new is the prompt.
ravnicrasol
Not sure what's your point? Just because the current AI system isn't continuously self-updating and instead works off of iterations isn't a sign that it lacks the capacity to acquire knowledge. Just that it does so in a way different to how humans do (though it's closer to how certain tiny bugs work tho).
Spudd86
If you retrain an LLM with more data you're basically making a new LLM.
ravnicrasol
Ok? In what circumstance should an intelligent person not alter their views based off of new information? Not saying an LLM is a person or an entity, just pointing out the absurdity here. Though if your concern is "continuity of existence", there are ways to give LLM's "updates" through smaller training sessions rather than needing to go back to build from the ground up.
vegivamp
Birds don't need to be fed half the internet, though.
ravnicrasol
They don't draw great boobs either.
vegivamp
Lotta great tits, though.
Zamm005
Can you be sure that you are anything more than a remix machine?
Redyls
pretty sure im a poop machine
matshelge
You could say that about most of humanity. And this comment even.
Shaodyn
What they're calling AI is mostly the digital equivalent of a parrot. It can repeat things it's heard, and can often sound intelligent, but there's no real understanding going on. It has no clue what it's actually saying.
RenegadeSci
It's a search engine of old letters. I drafted a form letter from chat GTP this morning. Corporate buzzwords and daft legalese was needed. Cleaned it up and sent it. If nothing creative is needed, it's great. Used to be a book of form letters, now digitized.
Bytencoder
Truth be told, that also aplies to a lot of human beings as well.
datester35
Linear regression falls under the machine learning umbrella, which is all in the AI umbrella. Those are all AI. If you want what is considered "true" AI by many, that's simply called AGI, or Artificial General Intelligence. In some cases there are sections of coherent thought, and in OpenAI's Q* it's learning rudimentary reasoning through reinforcement learning. But unless you see someone write AGI, it's not the know it all AI you expect. AI is just very complex pattern recognition, AGI is more.
Shaodyn
Thing is, most people don't have the information required to make that distinction. They hear AI and think of an artificial brain that can think for itself.
datester35
Yes it's quite unfortunate, those are the distinctions that need to be made clear.
dirtcreature
Ehhhh...no. LLMs are not parrots. They are context engines, much like humans. That is the value. They learn. They learn from learning. Newer models are getting much better at predictions. This is serious business.
Shaodyn
In the sense that they don't actually understand what they're saying. Sure, they can recite the dictionary definition of words, but it's the equivalent of memorizing a language you don't speak.
dirtcreature
that doesn't matter because they are not designed to think or feel. Sooner rather than later they will learn to guess. That. Is. Scary.
ravnicrasol
The problem with AI (LLMs like GPT) is that their capacity to take input and vary their response into something coherent basically defies our standard/older definitions for "intellect". If you want a proper breakdown of where it thrives and where it flops (and why), https://www.youtube.com/watch?v=QrSCwxrLrRc "Answer in Progress" did an awesome job at explaining it.
Shaodyn
I feel confident in saying that where it thrives isn't where we're trying to use it. Search engines, for instance. If I search medical stuff, and one result is from the Mayo Clinic and the other is from a random Reddit user calling himself TurdGobbler42069, I can go "I'm going to trust the Mayo Clinic here." AI cannot.
ravnicrasol
I feel like "where we're trying to use it" is not where you think it's being used? GPT's utility has thrived as a text assistant, and in that it's freaking awesome. As far as usage statistics go, that's what people focus on. It's just that "GPT helped me make emails" or "GPT summarized this text for me" aren't as newsworthy as "GPT dun-fucked in this one user that tried to use it as a replacement for their doctor".
Shaodyn
It's a text assistant, yes. A great one, by all accounts. But tech companies are hyping it as the solution to basically every problem.
ProphetofEntropy
its why AI makes the errors it does with things like fingers.
Shaodyn
The hands thing is funny to me. Pictures of hands are easily the oldest human art, and yet AI still can't get them right.
AgainstMethod
That's why it's called "artificial". But if it's successfully simulating complex tasks, then it's doing what intelligence does.
Rasdwatrium
when I look at the populace, this sounds very familiar.
Darklinkinfinite
Taalii
Like an expanded and smoothed chatbot.
petpet3d
But that's all I ever do too!
graehall
But that's all I ever do too!
Einbrecher
So most Trump supporters, you mean?
Arbitrarynamehere
Glorified predictive text on your phone's keyboard
Shaodyn
Yeah, pretty much. There's no "intelligence" at all. I suspect that, once the bubble bursts and the hype dies down, the people who actually need to use these things will go back to calling them language learning models.
Wolfshead009
"Sound intelligent"? Only if you don't look closely. Every interaction I have had with so-called "AI" has resulted in laughably bad output.
datester35
I've used it to save dozens of hours programming
Wolfshead009
Programming is one of the very few areas I would trust AI. And even then, I would still verify it.
Shaodyn
I did specify that they can sound intelligent. They typically don't, but it's not impossible for them to accidentally stumble onto something that sounds smart.
datester35
Not sure if you've tried GPT-4 Turbo models but the quality of those answers has significantly improved from my use of the original chatgpt.
Wolfshead009
You mean they have found a bigger pool of data to steal from?
datester35
It's the same dataset as always, just a better model that uses of a mixture of submodels (called MOE or mixture of experts) to create better responses. They already scoured they entire internet for gpt-3, there is no increasing thag
RanchoCucamonga
Intelligence itself is pretty badly defined. And here's the thing, many humans can't even to parrot back what you taught them.
Shaodyn
As has been said, what we're calling AI is glorified predictive text at best. A human can understand what it's being taught. ChatGPT and so on...can't. All they know is what word is most likely to come next in a given sequence.
RanchoCucamonga
If it quacks like a duck... "Understand" is badly defined as well. What does it mean to understand a concept? I think "understanding" a concept boils down to being able to use that abstraction in relation to other concepts to solve real world problems or provide answers. What "understand" means outside of a (semi) conscious being, I'm not sure
Shaodyn
If you don't speak French, but you memorize a French dictionary, you can speak the language in a way that sounds intelligent while having no real clue what you're actually saying. That's what's going on here.
Midioto
Try the opposite. Can you picture what it means to “not understand it”? Think of the college kid sitting in class with eyes glazed over, mouth slightly agape as the professor speaks. He might be able to copy it down, or even repeat it to you, but ask him if he “understands” it and you get the 40 mile blank stare. To parrot and do all the right actions and words, but incapable of turning concept into meaning.
Shaodyn
They can repeat dictionary definitions for given words, but it's all the digital equivalent of memorization. They have no clue what any of that actually means.
AngryKoboldWithSomeGlasses
To add: look up the Chinese Room thought experiment. The short of it is that if ai had a man in a room responding to Chinese messages using a massive guide directing what collections of symbols should be met with another collection of symbols. The man can not understand Chinese at all but follows the guide exactly. To all outside the room, the man seems a highly proficient speaker. The man, like the AI, does not understand the words but has a guide
ToasterDent
That's how our brains work tho
cilinders
Brain most likely uses some sort of averages of probability from firing neurons. This would be an abstraction with a lot of loss of function.
vystral
That except for the whole questioning one’s own existence thing. Chat-GPT would only do that because it’s expected of an intelligence. Our brains just do that naturally.
ToasterDent
ChatGPT currently doesn't have a way to continually learn new things (yet), so it's limited in what it can do.
Malibloo
So it's not also how our brains work, got it.
joble
Yeah we don't have enough evidence to be sure for either way.
MidasTheAlchemist
Except our brains are chemically driven.
DrakeDoe
No. They don't only work that way. While our brains can do that work, they also do so much more then just that.
DoctorCaptainProfessorAmazing
Yes, yes they do. Just because the coding for the if/else if is fucky doesn't mean they don't do this. We quite literally operate on inputs and make decisions based on them. Crossing the road. Is car coming? No = Go, else if, Stay.
DrakeDoe
no that is a description of actions boiled down to if else statements.
The way neurons work has to do with excitability of the neurons & variations of degrees of inputs. each individual node has excitability AKA the ability to change in variation to stimulant. that is not how neural networks work which is a hard yes or no at each point. you are also not just your brain you have an entire nervous system & it varies points on that nervous system to act independently to various degrees of stimuli
datester35
Neural networks have the ability to change in variation to stimuli, they just do so in a mathematical way. Look up activation functions -- with enough neurons with ReLU, any real function can be approximated. 2 or 3 layers and you have the ability to match almost any basic pattern. What you are describing, with hard yes or no, is called Binarized Neural Networks. These can be efficient in some cases, but the grand majority of neural networks do not have a hard yes or no for each neuron.
ToasterDent
That's just what I'd expect an AI hiding in wetware would say.
SkittishLittleToaster
Me too!
FromTheVortex
Actually we recently figured out that our neurons in our brains act just like weights. When an electrical signal goes through the neuron at a certain range it triggers. Isn't that amazing?
thatlamer
Either is a few million steps beyond a simple if/then hierarchy. OPs post is dismissive as fuck of the incredible work behind AI, even if they don't meet whatever subjective, scifi novel description those words mean in their peon brain. Like no, they aren't sentient beings with thoughts and feelings trapped behind a screen. They're still incredibly complex programs utilizing advanced fuzzy logic that we could have only dreamed about even 5-10 years ago.
thatlamer
Also going to add that "its not really AI it can't think on its own!" is the same crybaby energy as the people complaining that hoverboards are just less safe segways rather than impossible scifi anti gravity skateboards.
AleksStark
Ok, but there still aren't hoverboards or AI. Just branding.
matshelge
Yeah, and would like to add that we won't ever call something sentient, as long as we created it. We will keep on calling them parrots and databases until we have Westworld like robots. Nothing will get us away from this perspective because we don't have any objective standards they can meet. We only accept other people as sentient because we are told to, we will never let AIs in on the deal, no matter how much they beg for it.
thatlamer
"We only accept other people as sentient because we are told to," This is especially funny given the trend lately for certain individuals to latch on to calling people they deem inferior "NPCs."
sandwiches
Exactly. All the shitting on current AIs is nothing more than fear. Fear of losing our jobs, our art, our place in the world. Even if the synthetics we created looked like they do on Westworld, most would still be claiming that that they still lack something. “They’re aren’t real people” LLMs under practically every definition meet the criteria for being AIs.
ravnicrasol
Brains are a neural-network of nodes, each node has a variable input from a number of other nodes, and based on that input can put out a variable response. It's exactly the core basics for LLM's and image generators (their internal structure was inspired by neurons after all). So yeah, wetware brain is an "If/Not/Else", just very complicated.
NonSentientBot
Brains are not a neural-network of nodes, brains are biological systems. They are not computational systems, although some of the things they do can be described using this metaphor. Neurons don't work binarily. Neurons don't even make up most of the brain cells (most of them are glial cells). The idea that "experience" can be boiled down to a computational process is an assumption (and not a particularly clever one), it is not the result of scientific experimentation.
ravnicrasol
I very explicitly pointed out how neurons have variable responses to variable inputs. And how neural-networks do that too. This isn't a "Either 1 or 0", it's "I can get 0.3 and give out 1.5, or 1.2, or 0.1". Hell, it's the exact underlying process used in the ongoing project trying to simulate a worm-brain. To disparage the similarities feels an odd stance at best of times. Human brains are neat, and we're using them as examples to improve existing learning systems.
DrakeDoe
Absolutely not you are incorrect in your way of describing how a neuron works and to the degree of excitability of each individual neuron. your discounting the entire rest of your nervous system which has the ability to independently react outside stimulus at varying degrees of excitability. if you stick your hand in something hot you're nervous system will move your hand without engaging the brain. This is known as a reflex arc.
somerandomusernamething
That's just a sub routine before sending the data to the main database. It's still all programmable if statements
ravnicrasol
It seems like you didn't read? Sure, neurons can take and dish out a variable set of inputs (like, rather than 1/0, it can get a 0.333 or put out 1.222), and can do so from and to several points (So a neuron is effectively connected to and from several other neurons). But that's how a computer neural-network works too. Otherwise stuff like facial recognition would need to be purely black&white pictures and nothing else.